Anthropic has introduced a new 'auto mode' permissions setting for its Claude Code platform, aiming to strike a safer balance between restrictive defaults and the risky practice of completely skipping permissions. The default configuration in Claude Code is deliberately conservative, requiring approval for every file write and bash command execution. Previously, developers wanting to run longer tasks without constant interruptions often resorted to the dangerous workaround of disabling permissions entirely.

Auto mode offers a middle path, allowing tasks to run longer with fewer interruptions while significantly reducing risk compared to skipping permissions outright. Before each tool call, a classifier reviews the action for potentially destructive outcomes like mass file deletion, sensitive data exfiltration, or malicious code execution. Actions deemed safe proceed automatically; risky ones are blocked, prompting Claude to find an alternative approach.

If blocked actions persist, the system eventually triggers a user permission prompt. While auto mode reduces risk versus the 'dangerously-skip-permissions' approach, it doesn't eliminate it entirely, and Anthropic continues to recommend its use in isolated environments. The feature launches today as a research preview for Claude Teams users, with broader access for Enterprise and API customers coming in the coming days.